1 Feb, 11 tweets, 4 min read
Personal top 10 fallacies and paradoxes in statistics
1. Absence of evidence fallacy
2. Ecological fallacy
7. Prosecutors fallacy
8. Gambler’s fallacy
1. Absence of evidence fallacy

Absence of evidence is not the same as evidence of absence. Wouldn't it be great if not statistically significant would just mean "no effect"? bmj.com/content/311/70…
2. Ecological fallacy

Hard to resist those sweet population level data to make inferences about health effects on the individual level web.stanford.edu/class/ed260/fr…

If your goal is prediction, you may *not* be after unbiased predictor effects in your prediction model

Between group comparisons with baseline and follow-up: analysis of change scores or ANCOVA? Doesn't matter? Well it does...

errorstatistics.com/2019/08/02/s-s…

Perhaps one of the most famous paradoxes in statistics. Reversal of the direction of effect by simply combining two groups is something that may keep awake at night

Also known as collider bias, something we have seen plenty in the COVID-19 literature

nature.com/articles/s4146…
7. Prosecutor's fallacy

Pr(B|A) is not Pr(A|B). Confusing sensitivity/specificity for predictive values, p-values for probabilities about the hypothesis,.... the prosecutor's fallacy list is long

8. Gambler’s fallacy

Arguably the odd one in the list, but cognitive biases about probabilities of recurrent events are very real and relevant en.wikipedia.org/wiki/Gambler%2…

If you are interested in the Bayesian vs frequentist statistics wars, make sure you study the paradox

10. The low birth weight paradox

To adjust or not to adjust for birth weight in the analysis of infant mortality?

• • •

Missing some Tweet in this thread? You can try to force a refresh

This Thread may be Removed Anytime!

Twitter may remove this content at anytime! Save it as PDF for later use!

# More from @MaartenvSmeden

16 Jan
How to become a SUCCESSFUL academic: a guide 1/n
How do I know how to become a successful academic? I don't, but I have received plenty of advice. As a good academic, I will just summarize what I have learned from listening
1) Be the ultimate collaborator but also don't be

Say yes to as many collaborations as physically possible: co-produce papers, LEARN, co-write grants, DISCUSS, it is all about synergy. But also, collaborations slow you down, have your own ideas! Just say no to collaborations
15 Jan
The infamous retracted Hydroxychloroquire Lancet article?
Cited.... 883 TIMES
Only referenced as a joke or warning you say? Think again.. (screenshot from a 2021 paper)
At least the first author doesn't use it to boost his citations and H-index... oh wait...
21 Dec 20
Another year, another personal TOP 10 favorite methods papers
Disclaimer: this top 10 is just personal opinion. I’m biased towards explanatory methods and statistics articles relevant to health research, particularly those relating to prediction

The order in which the articles appear is pseudo-random
1) The first one is related to the pandemic. Title and subtitle give away the conclusions, but the arguments are particularly well put

science.sciencemag.org/content/368/64…
26 Oct 20
@Laconic_doc @statsmethods I think Alama has been called out by @GSCollins, I don't know about Public Health England.

@Laconic_doc @statsmethods @GSCollins That said, I have personally did quite a few things to warn you

First, I send you emails to which you politely and quickly responded. Thanks. You seemed to agree with my critique, but you didn't show any initiative to change it or remove the model
@Laconic_doc @statsmethods @GSCollins Second, I am one of the authors of a reply to the OpenSAFELY study where we specifically mention their model falls short of developing a risk model. You seem to have ignored that and used their multivariable results anyway
26 Oct 20
Today started with email with a new COVID mortality calculator send to a group of researchers

After contacting the developer they explained the calculator uses a *selection* of coefficients from multivariable models published in literature
they had no idea of predictive performance....

but acknowledges the limitations

are you kidding?
there is no doubt this "model" is meant to be used as a prediction tool and it is available online

acknowledging limitations is a really poor substitute for careful development and validation of what is essentially a medical device
14 Oct 20
HOW DO YOU DEVELOP A NEW PREDICTION MODEL?

This [THREAD] has been long in the making and is arguably overdue

1/138
I'll assume you have some basic knowledge of prediction models and will be relatively short on the technicalities

lets suppose you interested in developing a prediction model for disease X

2/138
There are probably a few dozen prediction models already developed for disease X!

most of them have never and will never be used

so... are you really, really, really sure the world is waiting for a new prediction model for disease X?

/138